Graduate Research

Permanent URI for this community

https://unbscholar.lib.unb.ca/handle/1882/12776

Browse

Now showing 1 - 8 of 8

Camera-LiDAR registration using LiDAR feature layers and deep learning
(University of New Brunswick, 2024-10) Leahy, Jennifer; Jabari, Shabnam
This thesis focuses on a new pipeline reducing registration error between optical camera images and LiDAR data, integrating the strengths of both modalities to improve spatial awareness. The first part presents an approach that enhances aerial camera-LiDAR correspondences through weighted and combined LiDAR feature layers comprising intensity, depth, and bearing angle attributes. Correspondences are attained using a 2D-2D Graph Neural Network pipeline and then registered using a 6-parameter affine transformation model, demonstrating pixel-level accuracies that improve its baselines. The second part introduces a new method for camera-LiDAR registration when the modalities come from different projection models, using combined LiDAR feature layers with state-of-the-art deep learning matching algorithms. We evaluate the SuperGlue and LoFTR models on terrestrial datasets from the TX5 scanner, and from a custom-made, low-cost Mobile Mapping System named SLAMM-BOT, across diverse scenes. Registration is achieved using collinearity equations and RANSAC.
Crack detection and dimensional assessment using smartphone sensors and deep learning
(University of New Brunswick, 2024-02) Tello-Gil, Carlos; Jabari, Shabnam; Waugh, Lloyd
This thesis addresses the critical challenge of deteriorating civil infrastructure due to natural processes and aging, emphasizing the importance of early detection for public safety. Surface cracks in concrete structures serve as vital indicators of deterioration, prompting the development of automatic defect detection using deep learning. Manual inspections, the basis of structural health monitoring, struggle with the complexities of crack patterns. The first part of this thesis focuses on training a Mask R-CNN network for crack detection, using augmented real-world data to enhance accuracy. The second part introduces a cost-effective methodology utilizing smartphone sensors' imagery and 3D data for automated crack detection and precise dimension assessment with YOLOv8 and Mask R-CNN. This research aims to advance a multi-modal approach combining LiDAR observations with image masks for accurate 3D crack measurements, establishing a pipeline for dimensional assessment, and evaluating state-of-the-art CNN-based networks for crack detection in real-life images.
Development of deep learning-based classification and unsupervised clustering methods for mineral mapping using remotely sensed hyperspectral data
(University of New Brunswick, 2024-12) Peyghambari, Sima; Zhang, Yun
Hyperspectral remotely sensed imagery is a powerful tool for mineral mapping. It captures detailed spectral information across hundreds of contiguous and narrow spectral bands to enable precise identification of various geological materials. Conventional methods mainly use shallow spectral absorption features to discriminate minerals and cannot extract their important spectral information. However, traditional methods face significant challenges in effectively handling hyperspectral data's high dimensionality, nonlinear spectral features, and low signal-to-noise ratio (SNR). These challenges limit the accuracy of traditional machine-learning algorithms in mapping the spectral variations of minerals. This PhD research addresses these limitations through a comprehensive literature review and the development of new methods. It has resulted in two published journal papers and one submitted journal paper, presented across three chapters of this dissertation. The third chapter of this dissertation (published review paper) provides an updated systematic overview of hyperspectral missions, diagnostic minerals' spectral properties, and various geologic information extraction techniques, including preprocessing, dimension reduction, endmember retrieval, and important image classification methods from spaceborne/airborne HSI. It evaluates the advantages and limitations of the existing conventional methods of processing HSIs with the aim of geological mapping. The fourth chapter (published paper) aims to improve the accuracy of spectral-spatial deep learning extractors in classifying HSI datasets. While traditional deep learning methods such as fully connected neural networks (FCNN), convolutional neural networks (CNNs), and hybrid CNNs like mixed convolutions and covariance pooling (MCNN-CP) algorithms have shown promise, they face limitations in robustness and accuracy. This proposes an integrated 1D, 2D, and 3D CNN architecture to enhance the capability of spectral-spatial extractors, significantly improving classification accuracy and resilience. The fifth chapter (submitted) explores deep learning-based clustering methods for unsupervised mineral mapping, which are valuable in remote areas where ground truth data is scarce. These methods leverage HSIs' high-dimensional and redundant spectral features, using advanced clustering techniques to generate accurate mineral maps without requiring extensive labelled data. This research proposes a hybrid 3D-2D convolutional autoencoder to capture HSI's spatial and spectral diversity. The anticipated outcomes include enhanced accuracy and computational efficiency, ultimately improving the utility of HSI for geological studies and resource exploration.
Development, testing, and ocean mapping application of three-dimensional baroclinic hydrodynamic ocean modeling at a variety of scales
(University of New Brunswick, 2024-12) Alleosfour, Ahmadreza; Church, Ian
To investigate the improvement in representing the general circulation and baroclinic condition of the Bay of Fundy and Minas Basin, two high-resolution 3D baroclinic hydrodynamic models have been developed using Finite-Volume Community Ocean Model (FVCOM). The effect of surface forcing resolution is assessed by prescribing two commonly used atmospheric forcing models in the region. The model simulations are validated against the available observations and collected Conductivity, Temperature, and Depth (CTD) data during a benthic habitat campaign in July 2018, the tidal constituents, Sea Surface Temperature (SST) data from satellite, and historical published current meter data. The models show a good agreement in capturing the tide both in the Bay of Fundy and Minas Basin and represented some improvement against the operational models in the region. In addition, the improvement in representing the baroclinic condition of the Bay both at the surface and water column is reported. The coarse prescribed surface forcing resulted in a warmer surface temperature in the Bay, while the higher resolution one generated a better agreement with the observations. Also, the model simulations have been implemented in an innovative approach to evaluate the partitioning of the hydrographic survey domain and provide insight into the multibeam echosounder depth uncertainty due to using a synthetized CTD profile, based on different criteria and limitations of hydrographic survey vessels for both Minas Basin and Saint John Harbor.
Forestry information extraction using high spatial resolution remote sensing imagery
(University of New Brunswick, 2023-12) Tong, Fei; Zhang, Yun
This PhD research focuses on the development of reliable and efficient methods for individual tree crown delineation and tree species classification, which provide essential information for modern forestry management and climate change monitoring. The primary objective of this research is to develop robust methods that can accurately delineate individual tree crowns and classify tree species using high spatial resolution remote sensing imagery. By achieving this objective, the research aims to enhance the reliability and efficiency of forestry management practices and contribute to the field of remote sensing applications in forestry. For tree crown delineation task, existing tree crown delineation methods are not suitable for large areas applications, because they need highly experienced experts to manually assign suitable parameters to control the delineation results, which is time-consuming, inaccurate, and not suitable for normal users. To address this issue, this dissertation presents a tree crown delineation method utilizing marker-controlled watershed segmentation specifically designed for high spatial resolution multispectral WorldView-3 satellite imagery. To reduce the difficulty of assigning parameters, the proposed method incorporates an automated supervised search process to determine the threshold. Moreover, an enhanced definition of spatial local maximum is employed to mitigate false treetops, thereby enhancing the accuracy of treetop detection. For tree species classification task, although deep learning methods based on convolutional neural networks (CNN) have achieved promising results, challenges remain in hyperparameter tuning and the requirement for large number of labeled training samples, limiting their applicability in real-world applications. This dissertation addresses these challenges by proposing three models based on the concept of deep forest to enhance the tree species classification from high spatial resolution hyperspectral imagery. All the three proposed models only require two hyperparameters that are easy to be determined by users. To optimize the classification accuracy, two different ways to combine both fixed-size patches and shape-adaptive superpixels are proposed to fully exploit spectral-spatial information within the high spatial resolution hyperspectral imagery. To reduce the demand for labeled training samples, the active learning (AL) is perfectly integrated into the multilayer cascaded random forests classification model.
Open-source web GIS for Arctic seafloor mapping: Improving the interactivity of public data dissemination
(University of New Brunswick, 2024-03) Vainionpää, Madeline; Church, Ian
The ease-of-use of online GIS software has encouraged many organizations to publish their geospatial data to a web mapping interface. With respect to the field of ocean mapping, this has been a breakthrough towards public accessibility since this data is easy to absorb in a visual format. There is a specific need for a Canadian Arctic-focused web portal which can host a unique dataset collected by the CCGS Amundsen over several decades. This project employs open-source JavaScript to not only display the Amundsen’s dataset on a web GIS interface, but to display it alongside third-party seabed data. The end user can manipulate the datasets using two interactive toolkits. The first is a statistical analysis toolkit which compares bathymetry raster imagepyramids that are hosted using web mapping services (WMS). The second is a three-dimensional visualization tool which virtually draws bathymetric WMS information in the same scene as cross-sectional seabed subsurface data.
Real-time Multibeam Echosounder error detection using deep learning
(University of New Brunswick, 2024-08) Chian Leal, Mary Oyuky; Church, Ian
The use of Uncrewed Surface Vessels (USVs) for marine surveying is increasing due to technological advancements, but they face logistical constraints of power and space availability. The autonomous nature of these systems would benefit from real-time detection of data errors using AI to enhance surveying capabilities. However, consideration for installing new devices capable of using deep learning algorithms on an USV must account for these constraints. Image segmentation is widely used in medicine to detect brain tumours and autonomous driving, and could be applied using the U-Net architecture to predict possible errors in real-time from USV Multibeam Echosounders. Tools like the Nvidia Jetson Orin AGX can facilitate real-time processing and analysis of data, while not impacting the operational efficiency of the USV. Integrating deep learning with USV operations shows promise in effectively identifying data errors, improving the automation of marine surveying, and simplifying data analysis.
The gravimetric geoid for Mexico: xGGM23
(University of New Brunswick, 2024-03) Avalos-Naranjo, David; Santos, Marcelo
The work presented in this thesis deals with the construction of a new gravimetric geoid model for Mexico. A large amount of terrestrial gravimetry collected up to year 2020 was processed in spectral combination with the satellite-derived geopotential model GOCO06s using the UNB’s Stokes-Helmert technique. The geoid model complies with international standards of the regional geoid for North and Central America and its resolution of 2.5 arc minutes is coherent with the actual spacing in gravimetry data holdings for the country. It was found that the new geoid model agrees with the national vertical datum by 10 cm in standard deviation, which implies a significant improvement from previous models. Improvements are mainly due to the recent densification of terrestrial gravity surveys, refinements in the software code of the SHGeo package, and an optimization process to select the frequency on which the terrestrial input data takes over from the satellite source.

Browse

Browsing Graduate Research by Subject "Geodesy and Geomatics"

Results Per Page

Sort Options